System Throughput Measurement on Embodied Rearrangement open-fridge (train)
2,861Mean SPSVER
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VERNumber of GPUs=8, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 2,861 | 3,829 | |
| NOVERNumber of GPUs=8, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 2,216 | 3,438 | |
| SampleFactoryNumber of GPUs=8, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 1,662 | 1,842 | |
| VERNumber of GPUs=4, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 1,432 | 1,915 | |
| SampleFactoryNumber of GPUs=4, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 1,286 | 1,568 | |
| NOVERNumber of GPUs=4, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 1,097 | 1,601 | |
| DD-PPONumber of GPUs=8, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 1,066 | 2,754 | |
| SampleFactoryNumber of GPUs=2, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 804 | 1,022 | |
| VERNumber of GPUs=2, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 716 | 945 | |
| NOVERNumber of GPUs=2, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 592 | 786 | |
| DD-PPONumber of GPUs=4, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 468 | 1,337 | |
| VERNumber of GPUs=1, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 428 | 534 | |
| SampleFactoryNumber of GPUs=1, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 427 | 517 | |
| NOVERNumber of GPUs=1, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 327 | 428 | |
| DD-PPONumber of GPUs=2, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 283 | 696 | |
| DD-PPONumber of GPUs=1, Hardware=Tesla V100(s), Environments per GPU=16, Training Steps=20 million2022.10 | 174 | 442 |